Noninvasive Computer Vision for “Tagging” Hawksbill Turtles

by Jason Holmberg and Colin Kingen

Beneath the surface of Hawaii’s blue ocean waters, Wild Me and Hawaiian Hawksbill Conservation are using photographs and computer vision technology to help protect Hawaiian hawksbill sea turtles. Photographs are commonly used to identify individual animals such as whales, zebras, and tigers for research and conservation purposes. Could they also be used to identify individual sea turtles?

sea turtle - Wild Me
photo by Wild Me

Typically, mark-recapture surveys involve physically tagging and later recapturing or resighting animals to determine distribution and movements and for developing population abundance estimates. The noninvasive method employed in this project would permit the study of these turtles using mark-recapture survey methodologies without the necessity of capturing, handling, and equipping turtles with physical tags attached to their shells or flippers. 

For the project, Wild Me and Hawaiian Hawksbill Conservation developed a computer algorithm that could search photographs of sea turtles to find matching individuals, satisfying the project’s goal of modernizing and speeding mark-recapture efforts for Hawaiian hawksbills to reliably and repeatedly identify individual sea turtles from photographs taken from various perspectives in their natural habitat. 

This innovative project involved sea turtle photographs collected by Hawaiian Hawksbill Conservation that were loaded into the Internet of Turtles (IoT) Wildbook platform (, allowing the computer vision system to automatically zero in on the turtles in the photographs. Once a turtle was detected, the IoT Wildbook for Hawaiian hawksbill sea turtles applied a trained deep convolutional neural network (technology similar to that used in facial recognition software) to focus on unique patterns on the turtles’ bodies and heads while determining the viewpoint (which side of the turtle it was seeing). The computer then searches other photographs to find matches, providing the researcher with a ranked list of potential matches.

Through other co-funding opportunities that integrated with the Christine Stevens Wildlife Award, Wild Me is now able to offer the capabilities of the IoT platform for hawksbill and green sea turtles to a global research audience. To date, the IoT is now supporting 36 users across the globe, who have entered 9,133 sightings of more than 3,700 individual turtles thus far. The Wildbook platform that the IoT is based on is designed to offer computer vision on a global scale to collaborating networks of researchers, providing a high tech, inexpensive, and humane platform to coordinate research efforts, especially across projects and borders. Thanks to the generosity of AWI, the IoT is a free platform that has moved beyond a concept and prototype and into a growth phase, offering time and cost-saving computer vision to noninvasively “tag” sea turtles in Hawaii and across the globe.

This study was funded by the Christine Stevens Wildlife Awards program. To learn more about this program or to view additional studies, click here.

Read more articles about: